Hi Mauro,
Because I answered to Caroline, I have to reply to your e-mail
> I'll be happy to hear our stat gurus on the contrasts proposed by
> Caroline.
>
>> For the 1st level analysis, the following t- contrasts are created:
>>
>> 1. [-1 1 0] for A vs Rest
>> 2. [-1 0 1] for B vs Rest
>> 3. [0 1 0] for condition A
>> 4. [0 0 1] for condition B
>
> My understanding was that, as all conditions are explicitely
> declared, contrasts 3 and 4 that are "conceptually" equivalent to 1
> and 2 should not work because, as they do not sum to 0, there is
> nothing left for SPM to compare them. I thought that 3 and 4 would
> work only if Rest had not been explicitely declared. Am I right or
> wrong?
Well I don't know for the SPM machinery if the fact to declare the rest
at the first level prevents contrast like 3 or 4 (but if you do it, SPM
doesn't insult you :-) ). In all cases, as mentioned in my reply to
Caroline, the use of the rest condition is better as it is explicitly
declared in the design matrix.
>> but what exactly are the differences between theses 6 comparisons?
>> How to do a correct ANOVA?
>
> If A and B are in the same run/session, there is maybe no need to
> compare them through Rest. You could compute A-B at the 1st level and
> take these con* files to the second level in a one-sample t test.
> (Right?)
Right! If you do (A-R) and (B-R) first level and next a paired t test at
the second level, it is the same as (A-B) first level and next one
sample t-test.
I've also proposed to Caroline a within subject ANOVA in my previous
e-mail (because she asked for), but it is clear for me that if it is a
real rest (e.g. nothing to do or just fixate a cross ..) a t-test is
more adapted.
Best,
cyril
|